In an attempt to surpass industry leaders ChatGPT and Gemini, Elon Musk’s AI-powered business, xAI, unveiled their frontier Grok 4 AI models with ground-breaking benchmark results.
Late on Wednesday, Elon Musk’s artificial intelligence startup, xAI, revealed SuperGrok Heavy, a new $300/month AI subscription plan, along with its newest flagship AI model, Grok 4.
Grok, xAI’s response to models like Google’s Gemini and OpenAI’s ChatGPT, can evaluate photos and provide answers. Grok has been further incorporated into Musk’s social network, X, which was just purchased by xAI, in recent months. But it has also brought millions of users’ attention to Grok’s misconduct.
During a webcast, xAI CEO Elon Musk revealed the flagship Grok 4 models, which include the Grok 4 and Grok 4 Heavy.
Grok 4 is expected to meet high standards. OpenAI’s upcoming AI model, GPT-5, which is anticipated to debut later this summer, will be compared to the most recent model from xAI.
Image analysis and advanced question answering are only two of the many state-of-the-art features included in the newly released xAI reasoning models.
During a webcast on Wednesday night, Elon Musk declared, “In terms of academic questions, Grok 4 is better than PhD level in every subject, no exceptions.” “It may not always have common sense, and it hasn’t yet developed new physics or technologies, but that will change in due course.”
Grok 4’s release coincides with a turbulent week for Elon Musk’s businesses. After about two years at the company, Linda Yaccarino resigned as CEO of X earlier on Wednesday. X has not yet named her replacement for the moment.
Yaccarino’s exit follows a few days after Grok’s official, automated X account praised Hitler and made antisemitic remarks in response to users who criticized Hollywood’s “Jewish executives.” Grok’s account had to be temporarily restricted by xAI, and the offending posts had to be removed. A newly introduced part to Grok’s public system prompt, which instructed the AI chatbot to refrain from making “politically incorrect” comments, seems to have been withdrawn by xAI in response to the event.
Musk and the leaders of xAI mostly shy away from talking about the incident, preferring to concentrate on Grok 4’s capabilities and performance.
Grok 4 and Grok 4 Heavy, xAI’s “multi-agent version” with enhanced performance, were the two models the business unveiled on Wednesday. According to Musk, Grok 4 Heavy generates several agents to work on a problem at once. The agents then compare their work “like a study group” to determine the optimal solution.
It is an interesting thing to note that Grok 4 achieved the highest score to date of 15.9% in the recently released ARC-AGI-2 benchmark.
According to xAI, Grok 4 performs at a frontier level across a number of benchmarks, such as Humanity’s Last Exam, a demanding exam that gauges AI’s capacity to respond to thousands of crowdsourced questions on topics including mathematics, the humanities, and natural science. Grok 4 outperformed Google’s Gemini 2.5 Pro, which scored 21.6%, and OpenAI’s o3 (high), which scored 21%, on Humanity’s Last Exam without “tools,” according to xAI, with a score of 25.4%.
According to xAI, Grok 4 Heavy with “tools” outperformed Gemini 2.5 Pro with tools, which scored 26.9%, with a score of 44.4%.
Grok achieves a new state-of-the-art score of 16.2% on the nonprofit Arc Prize’s ARC-AGI-2 exam, another challenging benchmark that comprises puzzle-like situations where an AI must identify visual patterns. That is almost twice as high as Claude Opus 4, the next-best commercial AI model.
Out of all the flagship AI models released by any AI lab, the model’s score was twice as high as that of Claude Opus 4 and OpenAI o3.
Grok has been quickly integrated into X (previously Twitter), the social media network currently controlled by xAI, in recent months.
In addition to Grok 4 and Grok 4 Heavy, xAI introduced SuperGrok Heavy, its priciest AI subscription package to date, which costs $300 per month. Plan subscribers will receive early access to new features and a first peek at Grok 4 Heavy. Although xAI currently offers the most costly membership among the main AI providers, the plan is comparable to the ultra-premium levels provided by OpenAI, Google, and Anthropic.
Some new products that xAI intends to introduce in the upcoming months may be made available to SuperGrok Heavy subscribers ahead of time. The business announced on Wednesday that it will release a video creation model in October, a multi-modal agent in September, and an AI coding model in August.
To encourage developers to create apps using the model, xAI is making Grok 4 available via its API. Although xAI’s enterprise sector is still in its infancy, the company intends to collaborate with hyperscalers to make Grok accessible via their cloud platforms.
It might be challenging for xAI to overcome its recent setbacks as it attempts to market Grok to companies as a serious competitor to ChatGPT, Claude, and Gemini, even while Grok performs at a frontier level on benchmarks. It’s unclear if companies are prepared to embrace Grok with all of its imperfections.
“With respect to academic questions, Grok 4 is better than PhD level in every subject, no exceptions,” the tech billionaire asserted during the broadcast.
He acknowledged that the model would continue to improve in order to reach perfection, but that it occasionally lacks common sense.
xAI unveiled SuperGrok Heavy, a $300/month membership service that grants access to the Grok 4 Heavy model.
We use cookies to ensure that we give you the best experience on our website. If you continue to use this site we will assume that you are happy with it.
